On Thu, Jul 16, 2026 at 11:08:55PM +0200, Jakub Jelinek wrote:
> I suspect this broke libstdc++ completely, the testing now takes
> significantly more time with parallel testing (and still pending).
> I normally do x86_64-linux and i686-linux bootstrap/make check in
> parallel, each with make -j32. The bootstrap times were today
> the same as yesterday, ~ 1h:10m for i686 and ~ 1h:30m for x86_64,
> but make check took yesterday ~ 1h:34m on i686 and ~ 1h:45m on x86_64,
> but today it is over 2h:40 on both and still nowhere the end.
> It seems as if instead of running each libstdc++ test once as before
> it now runs it 32 times (so each runtest.exp invocation runs it once).
Note, gcc-defs.exp contains
if { [info exists env(GCC_RUNTEST_PARALLELIZE_DIR)] \
&& [info procs runtest_file_p] != [list] \
&& [info procs gcc_parallelize_saved_runtest_file_p] == [list] } then {
and in there the using of a locking dir to decide which runtest.exp
invocation gets to execute which tests.
And so does libstdc++.exp, again
if { [info exists env(GCC_RUNTEST_PARALLELIZE_DIR)] \
&& [info procs runtest_file_p] != [list] \
&& [info procs gcc_parallelize_saved_runtest_file_p] == [list] } then {
with all that stuff.
So, I'm afraid that looks like the two can't be included together, either
one or the other (and not sure if the former can deal with the libstdc++
testing).
Jakub
